tansey / pycfr

A python implementation of Counterfactual Regret Minimization for poker
277 stars 70 forks source link

Do you obtain an low exploitability when using OS-MCCFR? #7

Open KK666-AI opened 6 years ago

KK666-AI commented 6 years ago

HI, i try to run your code 'test_oscfr.py' to obtain the exploitability curve for leduc hold'em, but i meet an error like that below:

next_reachprobs = [{ hc: reachprobs[player][hc[0:prevlen]] / possible_deals for hc in root.children[0].holecards[player] } for player in range(self.rules.players)] KeyError: (Ks,)

BTW, i write another code to calculate exploitability for leduc hold'em (stack=5), but i could only obtain an exploitability=0.4 after 100,000+ iterations. I am not sure it's the reason of the high variance of out-come sampling or there exists some bugs in my code.